Back
For best output, select "Paper Size" as "A4" and "Margin" as "0" or "None".
To save or print to PDF, please select Print Destination > Save as PDF, enable Background Graphics under "More Settings", then click "Save".
During the congress, E-Posters will be accessible to all participants on the congress website 24/7, as well as in the E-poster stations in the congress center.
Preparing your E-Poster
Please review the E-Poster format requirements carefully when preparing your E-Poster. Should your E-Poster not meet the mentioned requirements, it may not be displayed as described above.
E-Poster Submission Deadline
Please prepare and upload your E-Poster no later than March 14, 2026 11.59PM CET. After this date, you will no longer be able to prepare and upload your E-poster and it will not be displayed and accessible on the congress website.
Please follow the instructions below to input your abstract title.
Abstract titles should be brief and reflect the content of the abstract.
Acute kidney injury (AKI) represents a potentially life-threatening condition among hospitalized patients, with early prediction offering crucial opportunities for prevention. Despite advances in existing prediction models, clinical implementation has been hindered by excessive false positive rates (70%-99.4%) and inability to provide actionable clinical insights. This study aims to develop and validate a novel LLM-based framework that addresses these critical gaps.
We conducted a multi-center retrospective cohort study and developed two specialized large language models based on the Qwen2.5-7B architecture. The first model, AKI-PM, predicts AKI occurrence within 24 hours following pre-training on a specialized kidney corpus (4.26 billion tokens) and supervised fine-tuning. The second model, AKI-RAM, provides explainable risk attribution through a combination of fine-tuning and alignment techniques. We evaluated AKI-PM using standard discrimination metrics, while AKI-RAM underwent rigorous human evaluation to assess its clinical utility. The validation database comprised electronic health records from four geographically and institutionally diverse hospitals in China.
Our study included 140,637 hospital admissions from four independent institutions, allocated to fine-tuning (n = 47,750), internal validation (n = 17,074), and external validation (n = 75,813). The overall AKI incidence across all cohorts was 4.8%. AKI-PM outperformed six leading LLMs, achieving an accuracy of 0.91, AUC of 0.95, sensitivity of 0.79, specificity of 0.93, and positive predictive value (PPV) of 0.68 in internal validation. External validation demonstrated robust generalizability (accuracy 0.82-0.88; AUC 0.88-0.91; PPV 0.47-0.62), with further improvements using few-shot learning (accuracy 0.90-0.92; AUC 0.92-0.96; PPV 0.69-0.74). Performance remained consistent across most clinical subgroups. AKI-RAM effectively delivered structured risk attributions by categorizing modifiable and non-modifiable factors with tailored clinical recommendations. Clinical evaluation of 200 cases (50 from each of the four independent hospitals) by three nephrologists with more than 5 years of experience yielded high assessment scores (Likert 4.02-4.87) across eight clinical dimensions.
This study provides an accurate, interpretable, and scalable solution for AKI prediction and prevention, demonstrating strong potential for integration into diverse clinical environments to support early intervention and improve patient outcomes.
